Power and pitfalls of hierarchical clustering based asset allocation strategy in the Korean financial market

Abstract

The main idea of this study is to compare test investment strategies in Korean market using both traditional and hierarchical clustering-based allocation techniques: equal-weight portfolio, inverse variance optimization, minimum variance optimization, equal risk contribution, most diversified portfolio and single-linkage, complete linkage, average linkage and ward method portfolio from clustering algorithm. From the empirical result, there are advantages and disadvantages of clustered portfolio. Clustered portfolio generally avoids concentrated sector exposure, and without trading cost, out of sample risk-adjusted performance of clustered portfolio was better than that of traditional. Also, hierarchical structure on correlation matrix of individual assets makes economic interpretation feasible. Yet, the average turnover ratio per rebalancing is relatively higher than risk based asset allocation technique. In the time of crisis, correlation distanced based clustering tend to have less diversification effect. Compared to other algorithm, single linkage portfolio suffers from chaining problem producing concentrated portfolio when combining recursive bisection weighting scheme. Despite these disadvantages, diversification of portfolio is possible to add value through hierarchical clustering method as long as one is aware of its limitation.